debug

ultravox_model.py

@@ -201,6 +201,7 @@ class UltravoxModel(transformers.LlamaPreTrainedModel):
             for i, (audio, start, length) in enumerate(
                 zip(audio_embeds, audio_token_start_idx, audio_token_len)
             ):
+                print(f"{length=}, {audio.shape=}, {start=}, {audio.shape=}, {inputs_embeds.shape=}")
                 length = min(length, audio.shape[0])
                 inputs_embeds[i, start : start + length] = audio[:length]